Research map / Architectures and efficiency / CNN design
GhostNetV2: Enhance Cheap Operation with Long-Range Attention
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add DFC attention to Ghost modules (GhostNetV2) | GhostNet V1 · ImageNet val | top-1 accuracy: 74.5 -> 75.3 | Helped |
| GhostNetV2 backbone instead of GhostNet V1 | YOLOv3 · MS COCO 320 | AP: 21.8 -> 22.3 | Helped |
| GhostNetV2 backbone instead of GhostNet V1 | DeepLabV3 · ADE20K | mIoU: 34.17 -> 35.52 | Helped |
| add DFC attention | MobileNetV2 1.0x · ImageNet val | top-1 accuracy: 72.8 -> 75.4 | Helped |
| DFC attention instead of CA attention | MobileNetV2 · ImageNet val | top-1 accuracy: 74.5 -> 75.4 | Helped |
| add DFC attention on all stages (ablation) | GhostNetV2 1x · ImageNet val | top-1 accuracy: 73.9 -> 75.3 | Helped |
| apply DFC to expanded features rather than output features (ablation) | GhostNetV2 1x · ImageNet val | top-1 accuracy: 74.8 -> 75.3 | Helped |
| larger DFC kernel sizes than 1x3/3x1 (ablation) | GhostNetV2 · ImageNet val | top-1 accuracy | Helped |
| choice of up/down-sampling resize function (ablation) | GhostNetV2 · ImageNet val | top-1 accuracy | No clear effect |
| apply sigmoid before up-sampling rather than on full map (ablation) | GhostNetV2 · ImageNet val | top-1 accuracy | No clear effect |
Closest papers by meaning
- GhostNet: More Features From Cheap Operations (CVPR 2020 · similarity 0.89)
- MnasNet: Platform-Aware Neural Architecture Search for Mobile (CVPR 2019 · similarity 0.85)
- LSNet: See Large, Focus Small (CVPR 2025 · similarity 0.85)
- MobileODE: An Extra Lightweight Network (NeurIPS 2025 · similarity 0.85)
- FreeNet: Liberating Depth-Wise Separable Operations for Building Faster Mobile Vision Architectures (AAAI 2025 · similarity 0.84)
- IFORMER: INTEGRATING CONVNET AND TRANSFORMER FOR MOBILE APPLICATION (ICLR 2025 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.